Research And Implementation Of Multi-Label Marking System For Web Public Opinion Text

The Internet has become an indispensable social life style,it is not only sources of a large number of information,but also a hotbed of cyber crime.The events about Internet public opinion are happening with increasing frequency in recent years.The events about Internet public opinion,especially the events related to the people's livelihood,has serious influence on the prestige of the public management in China,also brought a lot of pressure to the public departments.Therefore,it has important practical significance for the stability of our society to take certain measures to positively guide the network public opinion and timely deal with those potential harmful factors.This thesis designs a multi-label marking system of network public opinion text according to the characteristics of network public opinion text by reference to the different functions and application methods of various domestic and overseas public opinion analysis systems.After text extraction,Chinese word segmentation,and stop words removal,the handled microblog text are classified and the classification results and other statistical data are shown on the web to complete data interaction between the system and the user,by using the characteristics of highly abstract feature,good fault tolerance and good adaptability of the convolution neural network.Eventually,the public departments could analyze the network public opinion's trend comprehensively,and grasp the characteristics of public opinion accurately,thereby achieving data mining.After deeply analysis of the topics of network public opinion,this thesis determines the classification system suitable for network public opinion text.The classification system includes six categories,namely,resources and environment,national security,social stability,government administration,daily life and individual behavior.They divide network public opinion text from different fields and levels.On the basis of this classification system,this thesis adopts the method of manual labeling to label the collected texts,providing reliable samples for the training and testing of convolutional neural network.After several adjustments to the network structure and parameters,the convolutional neural network suitable for multi-label classification of network public opinion text was successfully trained,and the expected effect was achieved.On the basis of the multi-label marking system of network public opinion text,the public departments could grasp the opinions and attitudes of hot issues and people's livelihood issues to formulate the feasible policies and measures to improve service levels,guide good faith public opinion,curb malicious rumors,improve their image and status,and popularize socialist core values.